A rule-based emotion-dependent feature extraction method for emotion analysis from speech

Area of Science:

• Speech processing
• Affective computing
• Linguistics

Background:

• Prosodic parameters in speech convey emotional information.
• Identifying emotion-specific vocal features is crucial for human-computer interaction and psychological studies.
• Existing methods may not fully capture the nuances of emotion-dependent speech characteristics.

Purpose of the Study:

• To develop a rule-based method for determining emotion-dependent features from high-level prosodic speech parameters.
• To quantify the similarity of emotional expression between speakers based on these features.
• To analyze similarities and differences in emotional speech expression across individuals.

Main Methods:

• A rule-based approach was employed to extract emotion-dependent features from statistical measurements of prosodic parameters.
• High-level features were selected using defined extraction rules.
• Emotional speech data from actors interpreting six emotions in four languages was utilized for evaluation.
• A ratio of emotional expression similarity was calculated based on the presence and values of identified features.

Main Results:

• All speakers exhibited universal signs related to certain emotion-dependent features in their emotional expressions.
• A significant finding was that nearly all speakers utilized unique sets and values for these emotion-dependent features.
• The analysis confirmed that emotional expression comparison can be based on feature sets and feature values.

Conclusions:

• The proposed method effectively identifies and quantifies emotion-dependent features in speech.
• While universal emotional expression cues exist, individual speakers demonstrate unique patterns in their vocal emotional delivery.
• Emotional speech analysis can be refined by considering both the specific features used and their corresponding values.